xref: /linux/arch/x86/kernel/vmlinux.lds.S (revision 9d4d8572a539ef807e21c196f145aa365fd52f0e)
1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 * ld script for the x86 kernel
4 *
5 * Historic 32-bit version written by Martin Mares <mj@atrey.karlin.mff.cuni.cz>
6 *
7 * Modernisation, unification and other changes and fixes:
8 *   Copyright (C) 2007-2009  Sam Ravnborg <sam@ravnborg.org>
9 *
10 *
11 * Don't define absolute symbols until and unless you know that symbol
12 * value is should remain constant even if kernel image is relocated
13 * at run time. Absolute symbols are not relocated. If symbol value should
14 * change if kernel is relocated, make the symbol section relative and
15 * put it inside the section definition.
16 */
17
18#ifdef CONFIG_X86_32
19#define LOAD_OFFSET __PAGE_OFFSET
20#else
21#define LOAD_OFFSET __START_KERNEL_map
22#endif
23
24#define RUNTIME_DISCARD_EXIT
25#define EMITS_PT_NOTE
26#define RO_EXCEPTION_TABLE_ALIGN	16
27
28#include <asm-generic/vmlinux.lds.h>
29#include <asm/asm-offsets.h>
30#include <asm/thread_info.h>
31#include <asm/page_types.h>
32#include <asm/orc_lookup.h>
33#include <asm/cache.h>
34#include <asm/boot.h>
35
36#undef i386     /* in case the preprocessor is a 32bit one */
37
38OUTPUT_FORMAT(CONFIG_OUTPUT_FORMAT)
39
40#ifdef CONFIG_X86_32
41OUTPUT_ARCH(i386)
42ENTRY(phys_startup_32)
43#else
44OUTPUT_ARCH(i386:x86-64)
45ENTRY(phys_startup_64)
46#endif
47
48jiffies = jiffies_64;
49
50#if defined(CONFIG_X86_64)
51/*
52 * On 64-bit, align RODATA to 2MB so we retain large page mappings for
53 * boundaries spanning kernel text, rodata and data sections.
54 *
55 * However, kernel identity mappings will have different RWX permissions
56 * to the pages mapping to text and to the pages padding (which are freed) the
57 * text section. Hence kernel identity mappings will be broken to smaller
58 * pages. For 64-bit, kernel text and kernel identity mappings are different,
59 * so we can enable protection checks as well as retain 2MB large page
60 * mappings for kernel text.
61 */
62#define X86_ALIGN_RODATA_BEGIN	. = ALIGN(HPAGE_SIZE);
63
64#define X86_ALIGN_RODATA_END					\
65		. = ALIGN(HPAGE_SIZE);				\
66		__end_rodata_hpage_align = .;			\
67		__end_rodata_aligned = .;
68
69#define ALIGN_ENTRY_TEXT_BEGIN	. = ALIGN(PMD_SIZE);
70#define ALIGN_ENTRY_TEXT_END	. = ALIGN(PMD_SIZE);
71
72/*
73 * This section contains data which will be mapped as decrypted. Memory
74 * encryption operates on a page basis. Make this section PMD-aligned
75 * to avoid splitting the pages while mapping the section early.
76 *
77 * Note: We use a separate section so that only this section gets
78 * decrypted to avoid exposing more than we wish.
79 */
80#define BSS_DECRYPTED						\
81	. = ALIGN(PMD_SIZE);					\
82	__start_bss_decrypted = .;				\
83	*(.bss..decrypted);					\
84	. = ALIGN(PAGE_SIZE);					\
85	__start_bss_decrypted_unused = .;			\
86	. = ALIGN(PMD_SIZE);					\
87	__end_bss_decrypted = .;				\
88
89#else
90
91#define X86_ALIGN_RODATA_BEGIN
92#define X86_ALIGN_RODATA_END					\
93		. = ALIGN(PAGE_SIZE);				\
94		__end_rodata_aligned = .;
95
96#define ALIGN_ENTRY_TEXT_BEGIN
97#define ALIGN_ENTRY_TEXT_END
98#define BSS_DECRYPTED
99
100#endif
101
102PHDRS {
103	text PT_LOAD FLAGS(5);          /* R_E */
104	data PT_LOAD FLAGS(6);          /* RW_ */
105#ifdef CONFIG_X86_64
106#ifdef CONFIG_SMP
107	percpu PT_LOAD FLAGS(6);        /* RW_ */
108#endif
109	init PT_LOAD FLAGS(7);          /* RWE */
110#endif
111	note PT_NOTE FLAGS(0);          /* ___ */
112}
113
114SECTIONS
115{
116#ifdef CONFIG_X86_32
117	. = LOAD_OFFSET + LOAD_PHYSICAL_ADDR;
118	phys_startup_32 = ABSOLUTE(startup_32 - LOAD_OFFSET);
119#else
120	. = __START_KERNEL;
121	phys_startup_64 = ABSOLUTE(startup_64 - LOAD_OFFSET);
122#endif
123
124	/* Text and read-only data */
125	.text :  AT(ADDR(.text) - LOAD_OFFSET) {
126		_text = .;
127		_stext = .;
128		/* bootstrapping code */
129		HEAD_TEXT
130		TEXT_TEXT
131		SCHED_TEXT
132		CPUIDLE_TEXT
133		LOCK_TEXT
134		KPROBES_TEXT
135		ALIGN_ENTRY_TEXT_BEGIN
136		ENTRY_TEXT
137		ALIGN_ENTRY_TEXT_END
138		SOFTIRQENTRY_TEXT
139		STATIC_CALL_TEXT
140		*(.fixup)
141		*(.gnu.warning)
142
143#ifdef CONFIG_RETPOLINE
144		__indirect_thunk_start = .;
145		*(.text.__x86.indirect_thunk)
146		__indirect_thunk_end = .;
147#endif
148	} :text =0xcccc
149
150	/* End of text section, which should occupy whole number of pages */
151	_etext = .;
152	. = ALIGN(PAGE_SIZE);
153
154	X86_ALIGN_RODATA_BEGIN
155	RO_DATA(PAGE_SIZE)
156	X86_ALIGN_RODATA_END
157
158	/* Data */
159	.data : AT(ADDR(.data) - LOAD_OFFSET) {
160		/* Start of data section */
161		_sdata = .;
162
163		/* init_task */
164		INIT_TASK_DATA(THREAD_SIZE)
165
166#ifdef CONFIG_X86_32
167		/* 32 bit has nosave before _edata */
168		NOSAVE_DATA
169#endif
170
171		PAGE_ALIGNED_DATA(PAGE_SIZE)
172
173		CACHELINE_ALIGNED_DATA(L1_CACHE_BYTES)
174
175		DATA_DATA
176		CONSTRUCTORS
177
178		/* rarely changed data like cpu maps */
179		READ_MOSTLY_DATA(INTERNODE_CACHE_BYTES)
180
181		/* End of data section */
182		_edata = .;
183	} :data
184
185	BUG_TABLE
186
187	ORC_UNWIND_TABLE
188
189	. = ALIGN(PAGE_SIZE);
190	__vvar_page = .;
191
192	.vvar : AT(ADDR(.vvar) - LOAD_OFFSET) {
193		/* work around gold bug 13023 */
194		__vvar_beginning_hack = .;
195
196		/* Place all vvars at the offsets in asm/vvar.h. */
197#define EMIT_VVAR(name, offset)				\
198		. = __vvar_beginning_hack + offset;	\
199		*(.vvar_ ## name)
200#include <asm/vvar.h>
201#undef EMIT_VVAR
202
203		/*
204		 * Pad the rest of the page with zeros.  Otherwise the loader
205		 * can leave garbage here.
206		 */
207		. = __vvar_beginning_hack + PAGE_SIZE;
208	} :data
209
210	. = ALIGN(__vvar_page + PAGE_SIZE, PAGE_SIZE);
211
212	/* Init code and data - will be freed after init */
213	. = ALIGN(PAGE_SIZE);
214	.init.begin : AT(ADDR(.init.begin) - LOAD_OFFSET) {
215		__init_begin = .; /* paired with __init_end */
216	}
217
218#if defined(CONFIG_X86_64) && defined(CONFIG_SMP)
219	/*
220	 * percpu offsets are zero-based on SMP.  PERCPU_VADDR() changes the
221	 * output PHDR, so the next output section - .init.text - should
222	 * start another segment - init.
223	 */
224	PERCPU_VADDR(INTERNODE_CACHE_BYTES, 0, :percpu)
225	ASSERT(SIZEOF(.data..percpu) < CONFIG_PHYSICAL_START,
226	       "per-CPU data too large - increase CONFIG_PHYSICAL_START")
227#endif
228
229	INIT_TEXT_SECTION(PAGE_SIZE)
230#ifdef CONFIG_X86_64
231	:init
232#endif
233
234	/*
235	 * Section for code used exclusively before alternatives are run. All
236	 * references to such code must be patched out by alternatives, normally
237	 * by using X86_FEATURE_ALWAYS CPU feature bit.
238	 *
239	 * See static_cpu_has() for an example.
240	 */
241	.altinstr_aux : AT(ADDR(.altinstr_aux) - LOAD_OFFSET) {
242		*(.altinstr_aux)
243	}
244
245	INIT_DATA_SECTION(16)
246
247	.x86_cpu_dev.init : AT(ADDR(.x86_cpu_dev.init) - LOAD_OFFSET) {
248		__x86_cpu_dev_start = .;
249		*(.x86_cpu_dev.init)
250		__x86_cpu_dev_end = .;
251	}
252
253#ifdef CONFIG_X86_INTEL_MID
254	.x86_intel_mid_dev.init : AT(ADDR(.x86_intel_mid_dev.init) - \
255								LOAD_OFFSET) {
256		__x86_intel_mid_dev_start = .;
257		*(.x86_intel_mid_dev.init)
258		__x86_intel_mid_dev_end = .;
259	}
260#endif
261
262	/*
263	 * start address and size of operations which during runtime
264	 * can be patched with virtualization friendly instructions or
265	 * baremetal native ones. Think page table operations.
266	 * Details in paravirt_types.h
267	 */
268	. = ALIGN(8);
269	.parainstructions : AT(ADDR(.parainstructions) - LOAD_OFFSET) {
270		__parainstructions = .;
271		*(.parainstructions)
272		__parainstructions_end = .;
273	}
274
275	/*
276	 * struct alt_inst entries. From the header (alternative.h):
277	 * "Alternative instructions for different CPU types or capabilities"
278	 * Think locking instructions on spinlocks.
279	 */
280	. = ALIGN(8);
281	.altinstructions : AT(ADDR(.altinstructions) - LOAD_OFFSET) {
282		__alt_instructions = .;
283		*(.altinstructions)
284		__alt_instructions_end = .;
285	}
286
287	/*
288	 * And here are the replacement instructions. The linker sticks
289	 * them as binary blobs. The .altinstructions has enough data to
290	 * get the address and the length of them to patch the kernel safely.
291	 */
292	.altinstr_replacement : AT(ADDR(.altinstr_replacement) - LOAD_OFFSET) {
293		*(.altinstr_replacement)
294	}
295
296	/*
297	 * struct iommu_table_entry entries are injected in this section.
298	 * It is an array of IOMMUs which during run time gets sorted depending
299	 * on its dependency order. After rootfs_initcall is complete
300	 * this section can be safely removed.
301	 */
302	.iommu_table : AT(ADDR(.iommu_table) - LOAD_OFFSET) {
303		__iommu_table = .;
304		*(.iommu_table)
305		__iommu_table_end = .;
306	}
307
308	. = ALIGN(8);
309	.apicdrivers : AT(ADDR(.apicdrivers) - LOAD_OFFSET) {
310		__apicdrivers = .;
311		*(.apicdrivers);
312		__apicdrivers_end = .;
313	}
314
315	. = ALIGN(8);
316	/*
317	 * .exit.text is discarded at runtime, not link time, to deal with
318	 *  references from .altinstructions
319	 */
320	.exit.text : AT(ADDR(.exit.text) - LOAD_OFFSET) {
321		EXIT_TEXT
322	}
323
324	.exit.data : AT(ADDR(.exit.data) - LOAD_OFFSET) {
325		EXIT_DATA
326	}
327
328#if !defined(CONFIG_X86_64) || !defined(CONFIG_SMP)
329	PERCPU_SECTION(INTERNODE_CACHE_BYTES)
330#endif
331
332	. = ALIGN(PAGE_SIZE);
333
334	/* freed after init ends here */
335	.init.end : AT(ADDR(.init.end) - LOAD_OFFSET) {
336		__init_end = .;
337	}
338
339	/*
340	 * smp_locks might be freed after init
341	 * start/end must be page aligned
342	 */
343	. = ALIGN(PAGE_SIZE);
344	.smp_locks : AT(ADDR(.smp_locks) - LOAD_OFFSET) {
345		__smp_locks = .;
346		*(.smp_locks)
347		. = ALIGN(PAGE_SIZE);
348		__smp_locks_end = .;
349	}
350
351#ifdef CONFIG_X86_64
352	.data_nosave : AT(ADDR(.data_nosave) - LOAD_OFFSET) {
353		NOSAVE_DATA
354	}
355#endif
356
357	/* BSS */
358	. = ALIGN(PAGE_SIZE);
359	.bss : AT(ADDR(.bss) - LOAD_OFFSET) {
360		__bss_start = .;
361		*(.bss..page_aligned)
362		. = ALIGN(PAGE_SIZE);
363		*(BSS_MAIN)
364		BSS_DECRYPTED
365		. = ALIGN(PAGE_SIZE);
366		__bss_stop = .;
367	}
368
369	/*
370	 * The memory occupied from _text to here, __end_of_kernel_reserve, is
371	 * automatically reserved in setup_arch(). Anything after here must be
372	 * explicitly reserved using memblock_reserve() or it will be discarded
373	 * and treated as available memory.
374	 */
375	__end_of_kernel_reserve = .;
376
377	. = ALIGN(PAGE_SIZE);
378	.brk : AT(ADDR(.brk) - LOAD_OFFSET) {
379		__brk_base = .;
380		. += 64 * 1024;		/* 64k alignment slop space */
381		*(.brk_reservation)	/* areas brk users have reserved */
382		__brk_limit = .;
383	}
384
385	. = ALIGN(PAGE_SIZE);		/* keep VO_INIT_SIZE page aligned */
386	_end = .;
387
388#ifdef CONFIG_AMD_MEM_ENCRYPT
389	/*
390	 * Early scratch/workarea section: Lives outside of the kernel proper
391	 * (_text - _end).
392	 *
393	 * Resides after _end because even though the .brk section is after
394	 * __end_of_kernel_reserve, the .brk section is later reserved as a
395	 * part of the kernel. Since it is located after __end_of_kernel_reserve
396	 * it will be discarded and become part of the available memory. As
397	 * such, it can only be used by very early boot code and must not be
398	 * needed afterwards.
399	 *
400	 * Currently used by SME for performing in-place encryption of the
401	 * kernel during boot. Resides on a 2MB boundary to simplify the
402	 * pagetable setup used for SME in-place encryption.
403	 */
404	. = ALIGN(HPAGE_SIZE);
405	.init.scratch : AT(ADDR(.init.scratch) - LOAD_OFFSET) {
406		__init_scratch_begin = .;
407		*(.init.scratch)
408		. = ALIGN(HPAGE_SIZE);
409		__init_scratch_end = .;
410	}
411#endif
412
413	STABS_DEBUG
414	DWARF_DEBUG
415	ELF_DETAILS
416
417	DISCARDS
418
419	/*
420	 * Make sure that the .got.plt is either completely empty or it
421	 * contains only the lazy dispatch entries.
422	 */
423	.got.plt (INFO) : { *(.got.plt) }
424	ASSERT(SIZEOF(.got.plt) == 0 ||
425#ifdef CONFIG_X86_64
426	       SIZEOF(.got.plt) == 0x18,
427#else
428	       SIZEOF(.got.plt) == 0xc,
429#endif
430	       "Unexpected GOT/PLT entries detected!")
431
432	/*
433	 * Sections that should stay zero sized, which is safer to
434	 * explicitly check instead of blindly discarding.
435	 */
436	.got : {
437		*(.got) *(.igot.*)
438	}
439	ASSERT(SIZEOF(.got) == 0, "Unexpected GOT entries detected!")
440
441	.plt : {
442		*(.plt) *(.plt.*) *(.iplt)
443	}
444	ASSERT(SIZEOF(.plt) == 0, "Unexpected run-time procedure linkages detected!")
445
446	.rel.dyn : {
447		*(.rel.*) *(.rel_*)
448	}
449	ASSERT(SIZEOF(.rel.dyn) == 0, "Unexpected run-time relocations (.rel) detected!")
450
451	.rela.dyn : {
452		*(.rela.*) *(.rela_*)
453	}
454	ASSERT(SIZEOF(.rela.dyn) == 0, "Unexpected run-time relocations (.rela) detected!")
455}
456
457/*
458 * The ASSERT() sink to . is intentional, for binutils 2.14 compatibility:
459 */
460. = ASSERT((_end - LOAD_OFFSET <= KERNEL_IMAGE_SIZE),
461	   "kernel image bigger than KERNEL_IMAGE_SIZE");
462
463#ifdef CONFIG_X86_64
464/*
465 * Per-cpu symbols which need to be offset from __per_cpu_load
466 * for the boot processor.
467 */
468#define INIT_PER_CPU(x) init_per_cpu__##x = ABSOLUTE(x) + __per_cpu_load
469INIT_PER_CPU(gdt_page);
470INIT_PER_CPU(fixed_percpu_data);
471INIT_PER_CPU(irq_stack_backing_store);
472
473#ifdef CONFIG_SMP
474. = ASSERT((fixed_percpu_data == 0),
475           "fixed_percpu_data is not at start of per-cpu area");
476#endif
477
478#endif /* CONFIG_X86_64 */
479
480#ifdef CONFIG_KEXEC_CORE
481#include <asm/kexec.h>
482
483. = ASSERT(kexec_control_code_size <= KEXEC_CONTROL_CODE_MAX_SIZE,
484           "kexec control code size is too big");
485#endif
486
487